ILLICIT DETECTION[ OVERVIEW ]
Regulated Bitcoin exchanges must screen every transaction for potential illicit activity (money laundering, sanctions evasion). This page compares two ML approaches: XGBoost (fast, explainable) vs GCN (graph-aware, catches network patterns).
ILLICIT TRANSACTION DETECTION[ MODEL_LOADED ]
Model Performance Comparison
| Metric | XGBoost | GCN | Winner |
|---|---|---|---|
| PR-AUC | |||
| Precision | |||
| Recall | |||
| F1 | |||
| Inference | ~2ms | ~50ms | XGBoost |
INTERACTIVE SCORING[ READY ]
Submit transaction features for real-time risk scoring with SHAP explanation.
THRESHOLD ANALYSIS[ INTERACTIVE ]
Adjust the classification threshold to see the tradeoff between false positives (manual reviews) and false negatives (missed illicit transactions).
MODEL CARD[ LOADED ]
illicit-xgboost
XGBoost classifier for illicit Bitcoin transaction detection. Uses 166 transaction-level features from the Elliptic dataset.
Last trained: 2026-03-01 06:06:05 UTC
Training Data
DATASET
Elliptic Bitcoin
N FEATURES
165
TRAIN SAMPLES
161721
TEST SAMPLES
29
TRAIN ILLICIT
157176
TEST ILLICIT
29
Training Config
model: XGBClassifier | n_estimators: 500 | max_depth: 6 | learning_rate: 0.1
Metrics
| Metric | Value |
|---|---|
| PR-AUC | 1.0000 |
| Threshold | 0.999 |
| Precision | 1.000 |
| Recall | 1.000 |
| F1 | 1.000 |
| Total Cost | $0 |
Pr Curve
Feature Importance
Confusion Matrix
Temporal
TRAINING PIPELINE[ PIPELINE ]
LOAD_DATA->TEMPORAL_SPLIT->TRAIN_XGBOOST->EVALUATE->SAVE_TO_R2